Market Twins, synthetic personas with real behaviour

The problem

Generic AI gives you the average. Market Twins give you your market.

Traditional research is slow, expensive and hard to repeat. Generic AI is fast but arbitrary: ask a language model the same thing a hundred times and you get a hundred different, random answers with no stable personality and no real spread.

We model the why behind human behaviour, psychologically grounded and scientifically validated. That is how decisions become reproducible and answers become real data without the noise.

Question: "Would you buy this product for 49 €?", asked 100×

Noise, every answer a different person Answers scatter randomly across the whole field. No stable pattern, no reliable signal.

The genesis of a twin

From a mathematical character to a human voice.

Step 01

A character is born

Every twin starts as a mathematical personality profile: values, risk appetite, price sensitivity, life stage. No prompt theatre, a stable core that persists.

Step 02

The twin decides, by inner logic

Ask a question and the decision layer decides first. Same question, same character, same answer, reproducible like a real person with conviction.

Step 03

A thousand twins, real spread

Across a whole sample, intended heterogeneity emerges: the frugal next to impulse buyers, sceptics next to early adopters. Diversity with structure, not dice in a mask.

Step 04

Only then does the twin speak

The language model only verbalises what is already fixed: it doesn't invent behaviour, it puts it into words. That is why the result sounds human and behaves the same way.

The Persona Engine

Human and twin, two layers, one logic.

The twin decides first and only then speaks. That makes the twin consistent like a person with a stable character, yet natural in expression.

Keep scrolling →

Decision layer

Behaviour you can compute

Personality, values and preferences are modelled mathematically. Every twin makes buying decisions by a stable inner logic, not a roll of the dice.

Stable character core across every repetition
Real, intended spread across the sample
Reproducible, analysable decisions

Language layer

A voice that invents nothing

The language model only verbalises what the decision layer has set: it doesn't invent behaviour, it puts it into words. That is why answers sound human and behave the same way.

Natural, context-rich phrasing
No hallucinating new preferences
Tone and stance matching the character

Use cases

One twin, many applications.

Two of many use cases, the same engine.

Market research

Market Twins

AI-moderated focus groups and consumer interviews with synthetic personas. Ask a question, survey a thousand twins, read the insights, in hours instead of weeks.

Product tests and concept screening
Pricing and willingness to pay
Audience segmentation in real time

Sales training

Sales Coach

Sales training on a synthetic lead. Your reps practise objections, discovery and closing against a realistic counterpart that never breaks character, before it counts with a real customer.

Onboard new reps in days
Objection handling in a safe space
Repeatable skill drills with feedback

Team

The people and the science behind the twins.

A team from market research, AI and science, with Prof. Marko Sarstedt as scientific advisor for methodology and validation. Not hype AI, but models we can prove.

Sandra Baethge

CEO & Co-Founder

Market research · Speaker & lecturer

Marcus Körner

Co-Founder

Market research & business

Ramtin Ramin

Co-Founder

Sales & scaling

Birger Krah

Chief Technology Officer

Platform & engineering

Christian Greb

Chief AI Officer

AI models & psychological models

Prof. Marko Sarstedt

Scientific Advisor

Methodology · LMU Munich

More about us and the science →

Live feel

Ask follow-ups. As deep as you like.

A static report answers one question. A twin answers the next one too. Lena, 34, price-conscious, sceptical of subscriptions, responds to every follow-up without breaking character.

Twin · Lena, 34 · price-conscious

LenaGo ahead, ask. I'm still Lena, however deep you go.

GDPR-compliant. By design, not an afterthought.

Built and hosted in Germany. Your twins, your data, your models, no leaking into foreign clouds, no compromise on compliance. This isn't a feature checkbox. It's the foundation.

Personas that think